An adaptive high performance architecture for "processable" bulk data transfers on a grid

For resource intensive tasks, effective utilization of resources is essential to achieve scalability and a satisfactory quality of service. The number of resources required and how the resources should be grouped depends on many factors, such as, network delays, the availability of resources and their current utilizations. Making such decisions at runtime using information, such as, current system state can improve performance. This paper proposes an adaptive architecture in which resource groupings and their patterns of interactions can be changed dynamically based on the task to be performed and the system state. A prototype implementation, valid for a certain type of tasks, is described. This architecture can be extended for other types of tasks using the concepts discussed in this paper